Letter pubs.acs.org/NanoLett
Terahertz Spectroscopy of Individual Single-Walled Carbon Nanotubes as a Probe of Luttinger Liquid Physics Joel D. Chudow,† Daniel F. Santavicca,†,‡ and Daniel E. Prober*,† †
Department of Applied Physics, Yale University, New Haven, Connecticut 06511, United States Department of Physics, University of North Florida, Jacksonville, Florida 32224, United States
‡
ABSTRACT: Luttinger liquid theory predicts that collective electron excitations due to strong electron−electron interactions in a one-dimensional (1D) system will result in a modification of the collective charge-propagation velocity. By utilizing a circuit model for an individual metallic single-walled carbon nanotube as a nanotransmission line, it has been shown that the frequencydependent terahertz impedance of a carbon nanotube can probe this expected 1D Luttinger liquid behavior. We excite terahertz standingwave resonances on individual antenna-coupled metallic singlewalled carbon nanotubes. The terahertz signal is rectified using the nanotube contact nonlinearity, allowing for a low-frequency readout of the coupled terahertz current. The charge velocity on the nanotube is determined from the terahertz spectral response. Our measurements show that a carbon nanotube can behave as a Luttinger liquid system with charge-propagation velocities that are faster than the Fermi velocity. Understanding what determines the charge velocity in low-dimensional conductors is important for the development of next generation nanodevices. KEYWORDS: Carbon nanotubes, Luttinger liquid, terahertz spectroscopy, one-dimensional conductors
C
behavior can be observed in the electrical-transport properties of carbon nanotubes. Another signature of the correlated behavior of electrons in 1D is the phenomenon of spin-charge separation. A 1D system supports collective excitations of charge and spin bosonic modes, which are separable and propagate at different velocities. The charge mode is commonly known as a plasmon, and the spin modes are referred to as neutral modes because they do not convey any net charge. The manifestation of spincharge separation via tunneling conduction has been experimentally observed for parallel nanowires in a GaAs−AlGaAs heterostructure18 and also in measurements of an electrostatically gated 1D system.19 The propagation characteristics of the charge-carrying plasmon excitations in 1D systems are dependent on the strength of the electron−electron interactions. In the present work, we utilize the frequency-dependent absorption of an individual metallic single-walled carbon nanotube to access this faster charge-mode velocity. The high-frequency properties of a single-walled carbon nanotube20 can be understood by treating the nanotube and ground plane as a transmission line. This model, first proposed by Bockrath21 and developed in detail by Burke,22 yields the transmission-line properties, including phase
arbon nanotubes are a prototypical experimental system for the verification of fundamental one-dimensional (1D) electron-conduction models and serve as attractive candidates for diverse potential applications. Due to their small diameter, the electron wave function is quantized around the circumference of a nanotube, resulting in an effectively 1D electronic system.1−4 Landau’s Fermi liquid theory breaks down in 1D due to weak screening of the Coulomb interaction. Instead, such a strongly interacting 1D electron gas forms a correlated state known as a Luttinger liquid, in which the low-energy electron excitations are collective bosonic soundlike density waves of spin and charge.5−7 A unitless order parameter g, known as the Luttinger parameter, quantifies the strength and nature of electron−electron interactions, with g < 1 for repulsive interactions and g = 1 for a noninteracting system such as a Fermi liquid. A hallmark of the Luttinger liquid model is a power-law dependence of tunneling conductivity on temperature and bias voltage. Theoretical work has predicted g ≈ 0.28 for metallic carbon nanotubes.8 Several reports on tunneling into ropes and individual carbon nanotubes claim agreement with this prediction.1,9−12 However, other possible explanations, such as dynamical Coulomb blockade, could result in the same observed behavior.13−15 Furthermore, experiments on multiwalled carbon nanotubes yielded similar results over an unexpectedly wide range of conditions.7,16,17 These results leave unresolved the question of whether Luttinger liquid © XXXX American Chemical Society
Received: April 10, 2016 Revised: June 28, 2016 Accepted: July 14, 2016
A
DOI: 10.1021/acs.nanolett.6b01485 Nano Lett. XXXX, XXX, XXX−XXX
Letter
Nano Letters
inductance LM ≈ pH/μm by several orders of magnitude, and so LM is omitted in the model. Due to the low density of states, the quantum capacitance is comparable in magnitude to the electrostatic capacitance CES, and both must be considered. CES between the nanotube and the environment, including the ground plane, is ∼50 aF/μm. CQ of each of the four channels is
velocity vp and characteristic impedance Zchar. This model is consistent with Luttinger liquid theory for describing the interacting electrons in a 1D system. The nanotube is modeled with the effective circuit given in Figure 1. Following Burke,22 included in this model is a kinetic
given by
2e 2 hvF
≈ 100 aF/μm. This value was experimentally
confirmed by Ilani et al.3 AC voltage excitations result in the dynamical charging− discharging of the nanotube, and it can be shown22 that these can accommodate four collective modes, one of which conveys charge, the plasmon or charge mode, and three that do not transfer any net charge, the spin or charge-neutral modes. The charge mode corresponds to currents through all the spin-up and -down channels in the same direction, and neutral modes are spin-up currents in one direction and spin-down currents in the other. This represents the separation of spin and charge as is expected in an interacting Luttinger liquid system. Because the neutral modes do not convey any charge, they do not interact through CES and will not couple to electric fields. We also include a term to account for damping due to electron scattering within the nanotube, denoted as an internal distributed resistance per unit length, rint. This rint for the best experimentally measured nanotubes is found to be ∼1 kΩ/μm at low temperature,25,26 which we assume for our simulations. In the circuit diagram, we incorporate the intrinsic quantum of resistance split equally between each end of the nanotube. The quantum of resistance of h/e2 = 25.8 kΩ for each quantum channel becomes RQ = h/4e2 = 6.45 kΩ for the four parallel channels of the nanotube.2,3 In addition to the quantum of resistance RQ, an additional contact resistance Rc between the nanotube and electrodes is included in parallel with a contact coupling capacitance Cc between the 3D metal electrodes and the nanotube. The contact resistance Rc arises from the
Figure 1. Equivalent circuit model of a carbon nanotube in a planar antenna geometry acting as THz source with impedance Rant. The enlargement highlights the effective transmission-line circuit diagram for the nanotube composed of an infinite series of resistances, inductances, and capacitances per unit length.
inductance, LK, accounting for the kinetic energy stored in the motion of the charge carriers, and a quantum capacitance, CQ, accounting for the finite band energy needed to add charge to the nanotube, as well as an electrostatic capacitance, CES, which is typical for a wire and a ground plane but also has a contribution due to electron−electron interactions. Carbon nanotubes possess two conduction channels from the orbital degeneracy of the two Dirac points in the unit sub-lattice, and each channel allows for spin degeneracy with electrons as either spin-up or spin-down.22 Thus, the complete model is composed of four connected transmission lines in parallel that interact with one another through their common connection to CES. LK is h2 ≈ 16 nH/μm per channel, corresponding to ≈4 nH/μm 2e v F
for the four parallel channels.22 LK exceeds the magnetic
Figure 2. (a) Predicted standing-wave resonances on a carbon nanotube in two limits of Rt compared to Zchar, with rint = 1 kΩ/μm. (b) Schematic of Fourier transform interferometer and antenna-coupled nanotube device for broadband THz generation and spectral response measurement. (c) Spectral response of niobium superconducting microbridge bolometers, with bowtie and log-spiral antenna structures showing the background system spectral response. B
DOI: 10.1021/acs.nanolett.6b01485 Nano Lett. XXXX, XXX, XXX−XXX
Letter
Nano Letters
the THz current that flows through the contacts. With this mechanism, we obtain an interferogram of each antennacoupled nanotube device in our Fourier transform interferometer setup. (The few samples that exhibited measurable bolometric detection had poor signal-to-noise ratios and did not produce meaningful interferograms.) The gate voltage, bias current, and temperature have a significant impact on the signal-to-noise ratio of the measurement.29 We employ considerable averaging, a low scan rate, long scan times, and zero padding of the interferograms30 by a factor of 2 to produce a clear spectral response. The spectral response is analyzed by attempting to identify the minima of power coupling or current flow, characterized by a dip in the spectral response (e.g., see Figure 3). Each dip in the THz response corresponds to a
imperfect transparency of the metal−nanotube interface; Rc is zero for a perfect contact. Even though the associated contact capacitance Cc is in fact a distributed capacitance between the electrode and the length of the nanotube, it is modeled as a parallel-lumped element here for simplicity. Calculating the wave velocity of the neutral modes, which do not include CES, gives vneutral = 1/ L k CQ , equal to the Fermi velocity, vF = 8 × 105 m/s. If repulsive electron−electron interactions are now accounted for with the inclusion of CES, the wave velocity for the four parallel channels is found to be greater than vF, vplasmon ≈
1 LK C 4 tot
4 1 + L K C ES L K CQ
≈
(1)
−1 where C−1 + C−1 tot = (4CQ) ES . The ratio of these two velocities is the dimensionless Luttinger parameter g, which quantifies the strength of the electron−electron interaction. The Luttinger parameter can also be determined via a capacitive relation composed of CQ and CES expressing the reduction of the electron−electron interactions by the screening of the electrostatic Coulomb force:
g≡
vF = vplasmon
C tot = 4CQ
C ES 4CQ + C ES
(2)
For a very weakly interacting system, CES ≫ CQ, Ctot → CQ, and g = 1. For a strongly interacting system, CES ≪ CQ, Ctot → CES, and g → 0. The characteristic impedance, Zchar =
LK C tot
=
LK 4C Q
+
LK C ES
=
1 h , g 8e 2
also varies with g. For
the four parallel quantum channels in an individual nanotube, g = 0.3 corresponds to Zchar ≈ 10 kΩ. The wave velocity can be determined by examining the power coupling to a finite-length mismatched transmission line, which acts as a quarter-wave impedance transformer.23 Plasmon standing-wave resonances occur when the nanotube length S corresponds to integral multiples of half-wavelengths. This vp λ situation occurs when S = n 2 = n 2f , where n is a positive v
integer, which gives resonant frequencies fn = n 2SFg with frequency spacing Δf =
vF 2Sg
Figure 3. (Top) Measured individual and averaged interferograms of 0.9 μm long carbon-nanotube device in bowtie antenna with side-gate using the nanotube contact nonlinearity as the THz detection mechanism. (Bottom) Normalized spectral response from Fourier transform of nanotube interferogram compared to a niobium bowtie background. There is a dip in the nanotube THz response at ≈0.6 THz, revealing the frequency spacing Δf. A simulation of the nanotube response from the transmission-line model, normalized by the system bandwidth determined by measuring a Nb microbolometer, reproduces this feature with a value of g = 0.7.
directly related to vp and g. For
Rt < Zchar, where Rt = Rc + RQ/2 is the total terminating contact resistance, input impedance minima occur at integer multiples of a half-wavelength, and input impedance maxima occur at odd multiples of a quarter-wavelength. For Rt > Zchar, the reverse is true, such that the input impedance maxima occur at halfwavelengths, and input impedance minima occur at odd multiples of a quarter-wavelength, as shown in Figure 2a. For the actual nanotube devices studied, Rt has a large range of values (20−500 kΩ) but is always in the Rt > Zchar regime. We previously reported measurements of rf and THz detection in individual carbon nanotubes, distinguishing between two response mechanisms: bolometric detection due to heating the nanotube electrons with a temperaturedependent resistance and the rectified THz response due to nonthermal electrical-contact nonlinearities.27−29 The nonthermal detection mechanism is dominant at low bias current. The nonlinear detection mechanism rectifies the THz current passing through the nanotube contacts. This rectification gives a DC response. Variation in the nanotube’s input impedance will modify the power coupling into the nanotube and, hence,
maximum in the input impedance. The frequency location of such a feature or the spacing of repetitive dips yields the value of Δf that is used to determine the charge propagation velocity vp and, hence, g. The equivalent circuit model used to represent the antenna as a voltage source can also influence the expected resonance behavior within the nanotube. The THz signal incident on the nanotube has an impedance approximately equal to that of the antenna (≈100 Ω), which is much less than the nanotube resistance (>10 kΩ), so it approximates a voltage source. Treating the antenna as an asymmetric or antisymmetric voltage source, and the relative distribution of the contact resistance on each end of the nanotube will affect the input C
DOI: 10.1021/acs.nanolett.6b01485 Nano Lett. XXXX, XXX, XXX−XXX
Letter
Nano Letters
the cryostat window eliminate signals at frequencies above the THz range. The system is maintained in a dry environment during measurement to minimize THz absorption by water vapor. For a nanotube in the FTIR system, features corresponding to standing-wave resonances appear in addition to the system spectral response. These periodic features (specifically, dips in the absorbed power) deviate from the antenna and optical-system spectral background and correspond to standing-wave resonances along the finite-length nanotube. The DC properties and the THz response as a function of bias current at low temperature for a ≈0.9 μm long nanotube in a bowtie antenna (sample NT-2) were previously reported.29 Using a bias current of 100 nA (low bias-current region) for nonlinear detection and a gate voltage of Vgate = −6 V, we measure an interferogram at low temperature (T ≈ 4 K), as shown in Figure 3. The gate voltage biases the quasi-metallic nanotube in the metallic state. The particular values of the bias current and gate voltage were chosen to optimize the signal-tonoise ratio of the rectified THz response. In addition to the average result of ≈100 scans, we also plot the individual data points during each scan to show the variation. Figure 3 also shows the normalized spectral response achieved from a Fourier transform of the interferogram average, as well as the standard error for each frequency data point from the collection of scans. Several other interferograms of this sample were taken at differing bias currents and gate voltages and exhibited the same behavior. There is a prominent dip in the spectral response at f ≈ 0.6 THz in comparison to the background response of the bowtie antenna and optical system. This background response is determined by measuring the bolometric response of a superconducting niobium (Nb) microbolometer with the same antenna geometry.40 The microbolometer has frequency-independent absorption over the measurement bandwidth and does not exhibit lengthdependent spectral-response variations. Assuming the nanotube has high-resistance contacts (Rt > Zchar), the dip at ≈0.6 THz corresponds to a half-wavelength resonance. At this resonance, there is an increased input impedance, which decreases the coupling of THz current to the nanotube and hence decreases the THz current through the nanotube contacts. The contacts partially rectify the THz current. The frequency of this dip is reproducible for the several measurements at different biasing conditions. Using this measured value of Δf ≈ 0.6 THz as the first resonance yields g ≈ 0.7, corresponding to vp ≈ 1.1 × 106 m/s. Higher-order resonances, which would be expected to occur at 1.2 THz (full wavelength) and 1.8 THz (three-halves wavelength), are beyond the bandwidth of the bowtie antenna and FTIR system. Simulations carried out using AWR Microwave Office of this nanotube as a transmission line with its measured length and total resistance, assuming rint = 1 kΩ/ μm,25 are used to determine the effective total capacitance per unit length, Ctot, that will produce an impedance maximum at f ≈ 0.6 THz. In Figure 3, we plot the product of this simulation and the normalized spectral response of the antenna system as measured with the Nb bolometer. We find good agreement between the measured nanotube THz response data and simulation with a value of Ctot ≈ 200 aF/μm, which is consistent with g ≈ 0.7. An uncertainly of ±0.05 THz in the experimental value of Δf would correspond to an uncertainty of ±11 aF/μm in Ctot and an uncertainly of ±0.06 in g. Using this nonlinear THz-detection approach, we have measured other individual antenna-coupled metallic single-
impedance of the nanotube and, thus, the magnitude of THz coupling into the nanotube. However, the resonant-frequency locations generally remain unmodified. For the special case of an antisymmetric voltage source with contact resistance divided evenly between the two ends of the nanotube, the even n resonances do not appear due to destructive interference on the nanotube transmission line. These resonances can reemerge as the difference between the contact resistances on each end of the nanotube increases. Also, with increasing values of Cc, the magnitude of each resonant feature increases as a function of frequency. In our model, we treat the antenna as an asymmetric voltage source and assume that the contact resistance is evenly divided between the two ends of the nanotube, each shunted by a small parallel Cc, as shown in Figure 1. Measurements of the dynamical conductance up to 10 GHz of single-walled carbon nanotubes with lengths over 100 μm did not observe any clear resonant behavior.24 This result was attributed to the damping of the plasmons by a large internal resistance, >100 kΩ, which dissipates the plasmon in a long nanotube and prevents the formation of standing-wave resonances. The scattering length due to electron−phonon interactions in a high-quality metallic single-walled carbon nanotube can be as a large as 10 μm at low temperature, and these nanotubes exhibit internal resistances per unit length of ∼1 kΩ/μm.25,26 Thus, short nanotubes, on the order of a few microns, measured at low temperatures are necessary to observe the predicted standing-wave resonance behavior. These shorter lengths necessitate higher measurement frequencies, ∼THz. Zhong and collaborators31 sought to measure vp using THz time-domain reflectometry of verybrief-duration electrical pulses in μm-long individual nanotube devices. They found resonances corresponding to propagation at the Fermi velocity. A faster plasmon velocity was not observed. Ngai and Kim32 theoretically considered this experiment and concluded that this experimental approach will excite all four modes available in a nanotube (three neutral spin modes and one charge mode), so the response may be dominated by the slower spin modes. They, like Burke, propose an alternate measurement scheme of vp based on coupling the charge plasmon mode to an applied oscillating electric field and observing AC conductance peaks versus drive frequency for μm-long metallic carbon nanotubes. Other theoretical treatments of the interaction of high-frequency radiation with carbon nanotubes to extract vp have been reported.22,33−35 THz detection and generation utilizing bundles of carbon nanotubes have also been reported.36,37 Plasmon resonances are predicted to be revealed in the THz thermal-emission spectrum of bundles of nanotubes, but no evidence of the resonances was observed.38 The use of heterogeneous bundles of nanotubes instead of a single individual nanotube may complicate and possibly prevent the observation of this behavior. To efficiently deliver THz radiation to an individual singlewalled metallic carbon nanotube, we utilize a quasi-optical coupling technique employing a lens and planar antenna geometry.39 The broadband THz generation and spectral response measurement is achieved using a Fourier transform interferometer (FTIR) with the nanotube mounted in an optical-access cryostat that is cooled with liquid helium,40 as shown in Figure 2b. The THz source is a silicon carbide globar (thermal source) heated to approximately 1300 K. The source is on−off modulated with a room-temperature chopping wheel, permitting an AC measurement of the low-frequency response to the THz power using a lock-in amplifier. Low-pass filters on D
DOI: 10.1021/acs.nanolett.6b01485 Nano Lett. XXXX, XXX, XXX−XXX
Letter
Nano Letters Table 1. Summary of Properties for Devices Discusseda contact structure
biasing conditions
sample length (μm)
antenna
gate
Ibias (nA)
Vgate (V)
R (kΩ)
Δf (THz)
g
Ctot (aF/μm)
0.9 1.1 3.2 4.6 3.1 3.2 2.6
bowtie bowtie log-spiral bowtie log-spiral bowtie bowtie
side-gate side-gate bottom-gate bottom-gate bottom-gate no gate no gate
50 30 5 20 10 100 3
−6 −3 −32 −0.6 −32
40 50 700 31 600 32 900
0.55 0.6 0.18 0.09 0.13 0.46 0.54
0.74 0.65 0.70 0.97 0.99 0.27 0.29
240 150 200 350 390 30 35
a
The measured DC resistance R includes the contact resistances, quantum resistance, and internal nanotube resistance. Also included is the value of Ctot determined from fitting simulations to the measured data.
Figure 4. (Left) Interferogram and normalized spectral response of a 1.1 μm long nanotube device in a bowtie antenna with a side-gate compared to the system background. Also shown are the data from the same sample measured in a different FTIR system showing the same dip in the THz response at ∼0.55 THz. (Right) Results for a 3.2 μm long nanotube device in log-spiral antenna with bottom-gate showing periodic features compared to a Nb log-spiral background and a simulation corresponding to g = 0.7.
Figure 5. Normalized spectral response of (a) a 4.6 μm long nanotube device in a bowtie antenna with a bottom-gate and (b) a 3.1 μm long nanotube device in a log-spiral antenna with a bottom-gate, both exhibiting periodic features corresponding to g ≈ 1. Results from (c) 3.2 μm long and (d) 2.6 μm long nanotube devices in a bowtie antenna with no gate electrode revealing slight features at ≈0.46 and ≈0.54 THz, both corresponding to g ≈ 0.3, with an assumed reduced antenna bandwidth due to optical misalignment.
E
DOI: 10.1021/acs.nanolett.6b01485 Nano Lett. XXXX, XXX, XXX−XXX
Letter
Nano Letters
misalignment of the sample in the FTIR system may be a significant contributor, such that the misalignment will cause shorter wavelength radiation not to intersect the antenna. For the simulation shown in Figure 5c, we assume a reduced bolometer antenna bandwidth to match the nanotube bandwidth envelope. Nevertheless, neither a larger contact capacitance nor misalignment explains the dip feature observed in the spectral response, which is well-fit as a first resonance corresponding to g ≈ 0.3. Other samples displayed no noticeable resonance features within the measured frequency range (possibly due to damping from a very large internal resistance), were too noisy to reliably determine any reproducible resonant features (often the case for higher-resistance samples), did not produce a well-defined interferogram (i.e., no central peak) despite reasonable nonlinear detection, or did not show any THz response at all. These samples highlight the challenges of obtaining uniform and robust properties in individual carbon nanotube devices. A comparison of the samples that displayed clear resonance features is shown on a single plot, Figure 6, in which the device
walled carbon nanotube devices of different lengths and antenna geometries. For each sample, the bias current and gate voltage were chosen to maximize the signal-to-noise ratio. Varying these parameters did not significantly modify the features in the spectral response, although the range over which the bias current and gate voltage could be varied was limited. Only bias currents corresponding to the zero-bias anomaly (typically ≲100 nA at ∼4 K) resulted in a clear THz response.29 Most samples required a minimum gate voltage to access the metallic regime. Many of these samples exhibited gate leakage at gate voltages beyond a particular value, and the presence of gate leakage drastically degraded the signal-to-noise ratio of the rectified THz response, effectively limiting the range of possible gate voltages. The results of the devices exhibiting resonance features in their spectral response can be classified into three general categories: (1) samples exhibiting a feature in the spectral response (as observed in NT-2) corresponding to g ≈ 0.7, (2) samples that showed a feature in the spectral response corresponding to g ≈ 1, and (3) samples that showed a feature in the spectral response corresponding to g ≈ 0.3. We briefly discuss representative samples for each category. Table 1 highlights the relevant properties and material parameters from the simulations for several of the devices studied. A second sample comparable in length and resistance to NT2, measuring ≈1.1 μm, at the center of a bowtie antenna with a side-gate shows similar behavior with a dip of ≈0.6 THz, as seen in Figure 4. This frequency also corresponds to g ≈ 0.7. This sample has also been measured in a similar FTIR system at the University of North Florida, revealing a dip in the spectral response at the same frequency. This reproducibility in two different experimental systems demonstrates the robustness of the spectral response. A longer nanotube, 3.2 μm, with a logspiral antenna geometry and a bottom-gate, shown in the inset of Figure 4, shows the periodicity of the dip feature in the spectral response. Measured at a bias current of IDC = 5 nA and Vgate= −32 V, the frequency spacing of the dip features for this device is Δf ≈ 0.17 THz, also corresponding to g ≈ 0.7. Simulations of this device agree with this value of g, as shown by the dotted line in Figure 4. A sample from the second category, for which we fit g ≈ 1, is a 4.6 μm long nanotube device in a bowtie antenna, measured at IDC = 20 nA and Vgate = −0.6 V, as shown in Figure 5a. Several other measurements at other bias conditions, such as 100 nA and Vgate = −10 V, show similar behavior but with a decreased signal-to-noise ratio. Figure 5b shows a 3.1 μm long nanotube device in a log-spiral antenna, measured at IDC = 10 nA and Vgate = −32 V also from this category. These 4.6 and 3.1 μm long devices, as well as a 3.0 μm bowtie antenna device and a 2.3 μm log-spiral antenna device, showed frequency spacing of 0.084, 0.13, 0.13, and 0.18 THz, respectively, all corresponding to g ≈ 1. A pair of the samples studied show features that correspond to lower g values, e.g., the sample shown in Figure 5c,d. The sample in Figure 5c is a 3.2 μm long nanotube in a bowtie antenna with no nearby gate electrode. It was measured at IDC = 100 nA and displays a slight dip within the maximum detection region at f ≈ 0.46 THz, corresponding to g ≈ 0.3. Similarly, the sample in Figure 5d is a 2.6 μm long device displaying a dip at f ≈ 0.54 THz. These devices showed limited bandwidth. If this reduction in bandwidth were entirely due to a large nanotube contact capacitance, it would require a value of Cc ≈ 500 aF, which is higher than the previous determination of the upper bound of the contact capacitance.29 Thus, optical
Figure 6. Plot of frequency spacing from spectral-response measurement as a function of nanotube inverse length comparing samples measured with identifiable resonance features. Several samples are best fit by a single-particle excitation given by g = 1, and two samples with no gate electrode are fit by g ≈ 0.3; a number of devices correspond to intermediate values near g ≈ 0.7. The antenna and gate geometries are specified in the legend.
antenna and gate geometries are also indicated. The slope of this plot of the measured resonant-frequency spacing Δf as a function of the inverse length of the device reveals the fitted value of g. A total of three distinct trends are evident. Several samples correspond to g ≈ 1, representing the noninteracting case with propagation at vF, consistent with the results obtained by Zhong et al.31 A pair of samples agree with the theoretical value of g ≈ 0.3. Half of the samples distinctly lie in between these two cases, with g ≈ 0.7. We conclude that single-walled carbon nanotubes can indeed exhibit a charge velocity that is faster than the Fermi velocity, as predicted by Luttinger liquid theory. Lee et al. reported on the imaging of 1D standing waves in metallic single-walled carbon nanotubes using scanning tunneling spectroscopy.41 They describe observing electronic standing waves produced by spin-charge separation and report that g ≈ 0.55. They explain the discrepancy between this value and the predicted theoretical value by stating that the underlying metallic substrate reduces the effective Coulomb interaction range of the electrons in the nanotube, increasing F
DOI: 10.1021/acs.nanolett.6b01485 Nano Lett. XXXX, XXX, XXX−XXX
Letter
Nano Letters the g value accordingly. This screening by the gate electrode42 is a possible explanation for the increased g value observed for many of our devices. As seen in Figure 6, the samples studied without a side-gate electrode show a g value closer to the predicted value of g ≈ 0.3. All of the samples that showed g ≈ 1 had bottom-gates. The samples with g ≈ 0.7 primarily had sidegates or no gates. This suggests that the proximity of a relatively large metal gate electrode is a source of electron screening that can modify the electron−electron interaction strength. Testing this further would be appealing, but it is difficult because a typical experimental quasi-metallic nanotube device often requires a gate to access the fully metallic regime. The select devices studied in this work without a gate were rare samples that naturally exhibited lower resistance. Future studies of nanotubes in varying contact conditions with alternative substrates, suspended geometries, or much smaller gates may further elucidate the effect of the electromagnetic environment on the charge-propagation velocity. Recent theoretical studies have examined the decay of a plasmon into the neutral spin modes within a carbon nanotube43 and have taken into account other higher-order processes that go beyond the simple Luttinger liquid picture.44,45 These processes can lead to a finite lifetime of the excitation, which would modify the total effective vp of the overall excitation. This may generate a more-complicated picture of the expected 1D electrical behavior46 and reevaluation of the predicted value for g. In summary, carbon nanotubes offer a laboratory in which to study 1D electron behavior, including the effects of strong electron−electron interactions as described by Luttinger liquid theory. One key prediction of this theory is that the fundamental low-energy excitations of the electron system are collective charge modes, or plasmons, that propagate faster than the Fermi velocity. We have studied individual micron-length antenna-coupled metallic single-walled carbon nanotubes in an FTIR spectrometer. We excite THz standing-wave resonances and measure the THz spectral response using the nonlinear nanotube contact resistance. The spectral response reveals standing-wave resonances consistent with a charge velocity vp that is equal to the Fermi velocity for several samples and faster than the Fermi velocity for others. The vp for some devices studied corresponds to a Luttinger interaction parameter of g ≈ 0.7, and a few devices agreed with the theoretically predicted value of g ≈ 0.3. We believe that the difference in the experimental values may arise from the different electromagnetic environments of the samples. A full understanding of this behavior may require that the theory be extended beyond the conventional Luttinger liquid picture. The experimental technique presented here offers a means of probing the correlated electrical properties of low-dimensional nanomaterials at THz frequencies. Understanding these properties is important for the design and realization of next-generation nanoelectronics based on 1D materials. Methods. The fabrication of antenna-coupled single-walled metallic carbon nanotube devices for THz measurement begins with the growth of carbon nanotubes using the chemical-vapordeposition technique on high-resistivity silicon (Si) substrates.47 The substrates have a top silicon oxide layer and, in some cases, a bottom metal gate electrode pattern beneath the oxide, which is approximately 300 nm thick (if a side-gate will not be used). High-resistivity Si is used because the THz radiation is transmitted through the back of the substrate, and charge carriers in doped Si would drastically attenuate the THz
signal. Isolated individual single-walled carbon nanotubes are located using scanning electron microscopy and atomic-force microscopy.25 We only select nanotubes with diameters of